yuting0624/antigravity-for-claude-code

What it solves

This project addresses the high cost and token-limit constraints of using a single LLM for complex software development tasks. It enables a "conductor-executor" workflow where Claude Code retains high-level judgment and verification, while delegating heavy, repetitive, or token-intensive tasks to a cheaper, faster Gemini-based agent (the Antigravity CLI).

How it works

It functions as a plugin for Claude Code that routes specific SDLC tasks—such as scaffolding, test generation, and deep research—to the Antigravity (agy) CLI. Claude acts as the orchestrator, reviewing and verifying the outputs generated by Gemini. The plugin supports features like cross-model verification, background delegation, and multimodal analysis (audio/video) without overwhelming Claude's context window. It uses a cost-aware routing policy to ensure that only tasks above a certain complexity threshold are delegated, maximizing efficiency.

Who it’s for

This tool is designed for developers who use Claude Code and want to optimize their AI-assisted development workflow for both cost and performance. It is particularly useful for those working on large codebases who need to perform bulk operations, internal data searches, or independent code reviews without incurring the full cost of running every step on a premium model.

Highlights

• Cost-Efficient Routing: Reduces frontier model costs by offloading bulk work to cheaper Gemini models.
• Cross-Model Verification: Allows Claude to act as a judge for work performed by a different model, improving reliability.
• Multimodal Capabilities: Processes audio and video files by generating timestamped digests, keeping Claude's context window clean.
• Native Integration: Commands appear directly in the Claude Code slash-command menu, providing a seamless user experience.
• Built-in Cost Discipline: Includes tools to measure and monitor token usage and cost-effectiveness across sessions.